CDI-Type I: A science and engineering programming and inverse modeling environment for massively parallel heterogeneous computing systems

该项目的目标是为科学家和工程师提供一个通用的、与硬件无关的编程和运行时环境，该环境利用了广泛的可伸缩的、高性能的、高度异构的硬件系统，同时对程序员隐藏了每个硬件系统的细节。该环境将进一步提供一个集成的逆建模工具，进行引导参数空间和不确定性分析以及模型优化。这样的逆向建模对于识别一般模式是至关重要的，以便开发表征复杂性的新科学理论，从而捕捉复杂的自然和工程系统的本质。除了提高我们对自然和构建系统（主要CDI主题）的复杂性的理解之外，所提出的环境将有助于可视化和从数据中提取知识（次要CDI主题），其中逆方法（例如，反卷积，回归，参数空间/不确定性分析，模型优化）是必不可少的。然而，只有将性能最高的异构硬件解决方案（可以执行数千次模拟）与高级编程和逆向建模环境（可以有效地隐藏异构硬件的复杂性）相结合，才能实现这些目标。确定了几个通用计算类，以及来自科学和工程的示例应用程序，它们可能受益于这种异构硬件实现。这样的实现将在科学和工程计算中开辟新的、转型的机会，增加发现新理论的可能性，这些理论涉及多尺度相互作用、紧急行为、模式形成和复杂的、倾向于反馈的系统中的自组织。一些基本的科学和工程计算类别，包括地震学、火山学、流体动力学和岩石磁学的应用实例，已经被确定为这种异质计算的目标。

The goal of this project is to provide scientists and engineers with a generalized, hardware-independent programming and runtime environment that takes advantage of the wide range of scalable, high-performance, highly heterogeneous hardware systems available, while masking the details of each from the programmer. This environment will further provide an integrated inverse modeling tool to perform guided parameter space and uncertainty analysis as well as model optimization. Such inverse modeling is critical to identify general patterns in order to develop new scientific theories that characterize complexity and thus capture the essence of complex natural and engineered systems. In addition to improving our understanding of complexity in natural and built systems (primary CDI theme), the proposed environment will aid in visualizing and extracting knowledge from data (secondary CDI theme) for which inverse methods (e.g., deconvolutions, regressions, parameter space/uncertainty analysis, model optimization) are indispensable. Reaching these goals is only possible, however, when combining the highest-performing heterogeneous hardware solutions that can execute many thousands of simulations with a high-level programming and inverse modeling environment that effectively hides the heterogeneous hardware complexities. Several general computing classes are identified, with example applications from science and engineering, that are likely to benefit from such heterogeneous hardware implementations. Such implementations would then open up new, transformational opportunities in scientific and engineering computing, increasing the likelihood of discovery of new theories regarding multi-scale interactions, emergent behavior, pattern formation, and self-organization in complex, feedback-prone systems.Several fundamental scientific and engineering computing categories with example applications from seismology, volcanology, hydrodynamics, and rock magnetics have been identified as targets of opportunity for this type of heterogeneous computing.